Question:
Are you putting together a Data Science Portfolio?
Asked on 2020-10-08 15:29 by Wojciech G.
On the October 7 webinar, we discussed the utility of having a portfolio with a few projects you can showcase or walk through. I'm curious if anyone is looking at building one out... If you'd like feedback, I'd be thrilled to hop on a quick call or chat and see what you're building or brainstorm how we can promote it!
Login to receive email notifications for this thread
Response #1By Greg L. on 2020-10-08 16:22I'm figuring this out. My approach is using GitHub+Medium for this, it feels a clunky. I see people sharing cool projects on LinkedIn (videos etc.).
Response #2By Leonard C. on 2020-10-08 17:01*I just noticed that a thread was started to discuss portfolios. I've copied my message from the feedback thread below...
Thanks for the informative webinar, Wojciech. I appreciated hearing your general input on DS employement and your perspective on how the industry and hiring process are evolving.
For those of us from academia (especially upcoming/recent graduates), having a "portfolio" to me seems like a tall task. From my perspective, such a process involves preparing a github repo with code and data, generating some example/representative results, and making the output readable, understandable, and exciting.
As an industry insider, what are some of the key elements from a portfolio that recruiters will be looking for? What's the best and most efficient way for those of us applying for jobs to spend our time?
Thanks!
Wojciech G. on 2020-10-08 17:20:The first thing I always try to ask myself is: how will a recruiter or hiring manager see this, and how do I make this as productive as possible for them?
Most managers will spend 5 minutes on an initial applications, which is unfortunately very short... They are seeing 100s or 1000s of applications, so how do you stand out?
I'd suggest that in your case, pick one or two papers you're particularly proud of and summarize those on a basic website (e.g., static GitHub page or Wix). Summarize it for a non-technical reader -- have 3 sentences that simply state what the innovation was.
If you're open to sharing a paper or two here, I can try to read it and tell you how you might frame it. melanie q. on 2020-10-10 04:48:Thanks. I didn't really think to write for a non technical audience. It makes me wonder how to find the right balance between the way I have to talk about my projects technically and non technically... Leonard C. on 2020-10-13 13:00:Wojciech, my response is a bit delayed because I was hoping to also link a github page with some content and code from the papers. For now, here are some papers I am interested in highlighting:
https://www.molbiolcell.org/doi/full/10.1091/mbc.E19-11-0614 (preprint @ https://www.biorxiv.org/content/10.1101/753681v1)
and
https://www.biorxiv.org/content/10.1101/2020.05.28.120907v1
Both papers include novelties in the data analysis techniques that I designed. For example, the first paper uses optical flow and a von Mises model (Figure 4D) to parameterize the distributions. The second paper involved more complicated image processing, and then a image-based resampling method similar to bootstrapping to assess randomness in the measured distribution.
Based on your input on the Oct 7th webinar, I am currently spending a significant amount of time making these tools available in github with example data, visualizations, explanations of the code, etc. Do you think this is the right way to continue? Here's the current repo (still in development) for the first paper:
https://github.com/ljcamp1624/OpticalFlowAnalysis
With some example movies, albeit with artificial data to test and debug the code:
https://github.com/ljcamp1624/OpticalFlowAnalysis/tree/master/AI1_2/MovieFolder
Hopefully it's clear that this is still in development and nowhere near the quality that I am envisioning for this repo. Thus, if this is in the right direction then I will continue pushing forward to make things cleaner.
Thanks! Wojciech G. on 2020-10-17 21:54:I think the work you've done here is very, very fascinating and definitely a great option for a portfolio.
That being said, I have to confess that it's even too scientific for me, and I think I tend to be more jargon-friendly than most folks who are hiring data scientists. How would you describe your findings in each paper in 1 or 2 sentences?
Response #3By Sara G. on 2020-10-09 03:23I was told to write a blog on medium to show I keep up to date with data science trends. I tried to write a blog and share this in my "portfolio" but honestly I struggled to find topics to write about. I worry that people would judge me for the way I write (it's just not my thing). I decided to create a private webpage to share with employers when I job search. I have found that I have to experiment a bit when coming up with ways to talk about my work.
Andrea Y. on 2020-10-10 04:25:I hear you...It took me a while to learn that having a blog meant that it was important for me to develop and share my perspective with others. It took me a lot of practice to feel comfortable with writing out my thoughts for the world...and sometimes I got mixed results, which is normal. Getting feedback from people who I trust and respect was really helpful in enabling me to find my voice. I also learned about how I wanted to write by learning from example aka other blogs that I enjoyed reading. melanie q. on 2020-10-10 04:46:Tried medium but the paywall was a real turnoff. Feels like there isn't a really great platform for really showing off my ML projects. I've watch a lot of youtube but not sure I'd want to be on video to explain my work lol Wojciech G. on 2020-10-13 23:57:A short Youtube video could be so interesting a portfolio choice. I've never considered that, but if you record a 1 or 2 minute video and play it as a project introduction and overview, that could be so powerful! Response #4By H L. on 2020-10-09 20:57For me, putting it into action is the biggest set back. I have a github with code from school work, and I techniques at work that I should reproduce with public data to put in my portfolio, but the sheer amount of work seems very overwhelming. From creating an article/blog/webpage to explaining everything, visuals code, seems like a big undertaking. I know the answer is to just do it....but it’s been a long and slow process.
Andrea Y. on 2020-10-10 04:20:I can appreciate the time and energy it takes to put this all together. I've found it helpful to time box this, meaning I give myself a few days to complete the task. One strategy I've found helpful is revisit my portfolio after a few months and make updates -- having had a chance to talk to employers, friends, colleagues, and others about my projects, I find that my message and key points become more concise, and it gets easier and easier for me to talk about. Another thing I've found helpful is to come up with consistent structure I use to tell my story (e.g., context, problem, solution; problem, reframe, solution, application; etc.) and stick with this. It helps people see patterns across my projects, and helps them following my thought process.
Good luck as you build it out. Perhaps you can share work in progress here or with others to get some feedback! Wojciech G. on 2020-10-11 01:24:HL, do you have examples of things you'd like to put into a portfolio? I'd love to check them out -- maybe we can help you figure out how to do it easily/quickly? H L. on 2020-10-11 04:02:Wojciech...
Most of what I have will either be school work (MNIST, housing price) which there are tons of already.
The alternative are analysis I do at work, which I can probably repurpose using public dataset. But then again, these analysis took a lot of effort (weeks and months) and EDA to model, so the question is, with a different dataset (and probably different domain), will it also take months to put together to make it "portfolio" ready.
I think the question I would have is this:
How "good" should a portfolio project be? does it need to have a high accuracy score? If I hack together a logistic regression model, tell the story, but dont take it further with fancy decision trees, variable selection, etc etc, is it any less impressive? I think the answer to this question is that it's about the story and also about the accuracy score, so it's both. :D
My personal approach (that i've been thinking about it for a long time) is to do it on a wordpress/square space type of thing and just have a consistent structure as Andrea has mentioned.
Wojciech G. on 2020-10-13 23:56:That's a good idea. GitHub's static pages are also pretty nice/simply and very cheap (if not free?) to run.
Not sure if you came to Jorge Escobedo's webinar talk, but his big piece of advice was to showcase a project where you collected data, cleaned it, analyzed it, and built a front-end... This sort of end-to-end approach is what he looks for. Response #5By Sarvesh S. on 2020-10-21 22:56Hi Wojciech, I would love to discuss my portfolio with you and get some feedback. I am building my personal website, I update my Github regularly and I try to showcase them on my LinkedIn as well.
Wojciech G. on 2020-10-21 23:11:Hi Sarvesh -- thanks for reaching out. Shoot us a note at hello@phaseai.com as we are running a few sessions on portfolios, where people get feedback from us and each other. The next one is on Oct 23 Friday at 2pm Eastern Time, though it'll be full soon. We'll have others next week. You can also see a recent post
we wrote on the topic.
Sarvesh S. on 2020-10-22 15:19:Thank you, I sent an email at hello@phaseai.com
Response #6By Leonard C. on 2020-11-05 14:28It's taken a while since our Oct 23rd meeting, but here's a recent update on my "portfolio" building: I've built a personal website that can be found at
ljcamp1624.github.ioI would love to get some input on my website, and I'd also be happy to provide feedback to others currently working on a portfolio and/or a website!
Andrea Y. on 2020-11-06 00:29:First off, way to update your portfolio! It looks awesome and I know how hard it is to put these things together. The visuals really make your portfolio stand out and help your projects come to life.
I have a few points of feedback you may want to consider...and I welcome your thoughts!
- Tell us more about the way you think in your bio: If you have a point of view on what you do/how you do it, share it here. What sets you apart from other people who do similar work? What does your approach look like? Why is your work important? Also, I'd suggest leading with your expertise first rather than your education.
- Use a consistent structure to describe your projects: You've done a great job of describing each project. For the most part, the current structure looks like: title, description (what/how), featured publications, and key concepts. I'd encourage you to use this consistently so the reader knows what to expect and can compare across projects. A structure that I've personally used in the past and find helpful is: Role (my role in the project), Goal, Approach, and Insights to Action (outcomes/implications). For all projects, I'd suggest you add the "insights to action" part -- What were the outcomes/consequences/applications of your work? Why does this make a difference to other researchers, businesses, decision makers, etc.? Why should someone care?
- Create white space: When it comes to layout, I'd suggest creating a bit more white space between your content pieces. Right now, it feels like the text runs into each other, and it makes it harder to distinguish between your projects. Same goes for publications, give some spacing so it's easier to see the list.
Hope that helps a little! I think it looks awesome.
Sara G. on 2020-11-07 19:35:Cool portfolio! Your work sounds so interesting!
One thing that I've found helpful is have a longer version of some of my portfolio pieces. So for instance, being able to click into a new page that has a longer and more detailed explanation of my work and links to publications, etc. These are the projects I may bring to an interview to share with the person who is interviewing me so I can talk them through the project and show them multiple images/models I've created.
Vasundhara V. on 2020-11-25 20:08:Hello Leonard
Very nice portfolio. I've been trying to make one as well, but I'm not very good with website related languages. Did you use a Javascript for the simulations? Could you tell me how you did that? I have been trying to add some previous simulations I did (Simulink and Java applets) but I am unable to add them to a website.
Thanks
Vasu
Rachel W. on 2020-11-25 23:45:Hi Vasu! Another approach could be to create a short (1-2 minute) video presentation on what you did. We're finding this to be more common, especially if you aren't a web developer. We can help you get feedback if you want to try something like this, just let us know.
Vasundhara V. on 2020-11-27 18:24:Hi Rachel
Thank you so much. That would be nice. I have graphs and figures for 1-2 important projects, which is good. But there's 1 project for which I want to add simulations, so are you suggesting I make a video for these simulations and add, instead of adding the simulation? Could you recommend some platforms of making videos like this?
Thanks
Vasu
Andrea Y. on 2020-11-28 02:13:Hi Vasu! Rachel was actually referring to creating a video of yourself talking about your project. We're testing a tool to help people share their projects/portfolios with potential employers. Video is the primary way to communicate these days, plus we've noticed a growing interest from employers for videos like this from job seekers.
If you're open to making a 30sec recording about a project that you're working on to help us test the process/software, we'd be more than happy to provide you with feedback on your video pitch. You can keep a copy of the recording for your own purposes.
The software isn’t meant to be polished at this point. Requirements: Phase AI account + Chrome or Firefox on a desktop/laptop (not mobile). If you have any questions/issues, feel free to reach out to me at a@phaseai.com.
Thanks,
Andrea